发送短信: On guaranteed optimal robust explanations for NLP models